The Chinese government would strategically prefer companies were not using Nvidia chips, since not doing so accelerates the domestic alternative. All that is officially allowed from Nvidia, on a case by case approval basis, are older Hopper-generation H200 chips, and these are what the Chinese government are themselves blocking.

The Chinese AI companies themselves are constrained by the supply of the domestic alternative (primarily Huawei's Ascend series), as well as domestic HBM supply, and appear happy to use whatever they can get their hands on, the faster the better (Megaspeed have been buying newer Blackwell chips), or have API access to.